Considering the trend of small hydropower towards low head, and reusing the water surplus head of cooling tower, a new type of turbine with small size, high efficiency and super low specific speed was proposed to be as direct drive for fan in large cooling tower or applied in small hydropower plants for generating electricity, which was designed and validated based on design theory of fluid mechanics and validation method of numerical simulations. Combining binary theory and spiral potential flow to design the flow streamline in volute, the new type volute adopted axial outlet to ensure the turbine radial dimension was about half compared with similar conventional turbines, this kind of structure was a great benefit to decrease manufacturing cost, which was more conducive to ventilation in cooling tower. The work of impact annular blades reduced the turbine specific speed to a large extent which adapted to the internal structure of the cooling tower, the tangent of runner blade was perpendicular to the runner axis. Corresponding to the runner outlet, draft tube was designed with annual inlet, the water flow through the four water distribution pipes in the draft tube which was propitious to improve flow field performance in draft tube, meanwhile, the turbine was able to mount on the central base directly with this structure. Theoretical calculation on hydraulic loss in each domain was given and it was compared with corresponding data that acquired from numerical calculations;the results showed that the hydraulic loss in each domain was relatively small, and differences between them were less than 5%. Numerical calculations were carried out with model which was built in Solidedge and meshed in ICEM, SST k-ω was used in all simulations to capture fluid details, while monitoring points were located on guide vane and runner blade for obtaining pressure values. Numerical results showed that velocity at the volute outlet was consistent with constant velocity moment law and the flow field characteristics illustrated this model with good performance, and the predicting efficiency was around 90%.